Process Modeling Leveraged by Workflow Structure and Running Logs Analysis

Abstract

The reality of big data opens up a new world for business process modeling. Omnipresent cases and workflow logs are getting accessible, which implies the chance to exploit important patterns hidden in them so as to cut down the modeling cost or to improve the quality of process models. To take the full advantage of big data in process-aware systems (PASs), we propose a novel business process modeling technique that leverages the modeling by cases and workflow logs analysis. It uses the average perceptron to analyze both of existing process structure of cases and co-occurrence relation of activities in workflow logs. In contrast to traditional manual efforts, it improves the performance significantly by recommending proved working patterns. Comparing to recent process mining strategies, it serves the modeling online with meaningful process segments. We evaluate our approach against a synthesis dataset (100 processes and 10,000 log items generated by the plugin PLG in ProM) and real data from public business processes (77 processes in the package Paul Fisher workflows for benchmarks PR and CA2 from the website myExperiment). The study reveals that 9.46 % improvement in precision can be gained by considering both case structure and log items in contrast to the structure only, or 5.94 % gaining in contrast to mere logs. Our evaluation validates the effectiveness of the proposed technique and efficiency when we applying it on real modeling scenarios.